Hey there!
Summary of this email:
- 🗞️ News of the week: Opus 5.5 has Fable quality and costs 40% less than Opus 5
- Kafka can now have more than one consumer per partition!!! (check it out in the course)
- Café con Codely recap (watch it here)
- The joke of the week
🗞️ News of the week: Opus 5.5 has Fable quality and costs 40% less than Opus 5
Last Tuesday Anthropic released Claude Opus 5.5.
We have used it quite a lot, and these are our first impressions:
- Opus 5.5 on medium gives results similar to Fable 5.1 on low.
- Its answers are much clearer.
- It uses far fewer tokens. In Claude Code,
/usagebarely seems to move with Opus 5.5. - It is faster.
- It has better taste for design.
Also, if you use the API, they have lowered the price.
It drops to $4 per million input tokens and $20 per million output tokens, compared to $5 and $25 for Opus 5. Cache reads also drop from $0.50 to $0.20.
According to Anthropic, on typical workloads it costs 40% less than Opus 5 and generates output more than 30% faster.
And on the benchmarks Anthropic published, Opus 5.5 beats Fable 5.1 on agentic coding, computer use and knowledge work. A smaller and cheaper model beats the big one.
Tomorrow we will be live at 9 CEST discussing this story and many more on Café con Codely. On our YouTube, Twitch and X.
📯 Kafka can now have more than one consumer per partition
Historically Kafka has had a limitation: at most, one consumer per partition.
That is why the partitioning strategy of the topics was so critical. If you wanted more consumers, you needed more partitions.
With Kafka Queues this is no longer the case. Several consumers of the same share group can read from the same partition. If the partition "piles up" because a lot of events arrive, you can start more consumers.
And you do not have to coordinate anything, you can delegate it all to Kafka.
Kafka also distributes the work by itself. With two partitions and two consumers, the automatic rebalance decides that each consumer takes one partition, without you saying who goes where.
We cover all of this with code and with a share group visualizer in the Event Bus in Kafka course, together with retries, dead letter and how to extend the timeout of a message with renew.
Now 100% published on Codely Pro Premium in early access. We hope you enjoy it as much as we enjoyed making it. 😊
☕ Café con Codely recap
This is the top 3 of headlines we discussed in the latest Café:
- Jev: the model that does not generate text and picks between options 20 to 200 times faster than frontier models. In the invoice demo it picks between fraud, clean or manual review with 88% confidence, while an LLM takes 8.5 seconds to write its answer.
- Dario Amodei: asks to slow down AI and, for once, Altman and Musk agree with him. His essay proposes external evaluators inside the labs, agreements between democratic countries and, at the end, global coordination with China.
- Shopify: leaves React Native and goes back to Swift and Kotlin because agents make it cheaper to build twice. To avoid slop in the migration it built Helix, which compares each screen with the native version through checkpoints.
You have the full Café recap with links to each story at the moment we discussed it.
Tomorrow we will be live at 9 CEST discussing this week's news on Café con Codely. On our YouTube, Twitch and X. 🙌
And since you've made it to this part of the newsletter, here's the joke of the week, which I know you were waiting for:
Which font always tells the truth? Times New Roman 12! It's the one lawyers use for legal papers, so everything in it is legally binding… 😂 😂 😂
Cheers!